Reproduce “LoopedWAM: Looped World Action Models for Elastic Inference” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Anh Dao, Xinxin Liu, Hang Yu, Jeffrey A Chan Santiago, Yuzhang Shang
LoopedWAM is a robot policy that maps camera images and instructions to actions while reducing stored blocks through shared transformer cores. It trains an elastic student for different action-loop and sampling-step budgets, while generating future video only during training. On LIBERO, it reaches 96.0% success in 27.6 ms at two loops and steps, but remains weaker than Fast-WAM on LIBERO-Long.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.